220 likes | 345 Views
Processing of large document collections. Part 7 (Text summarization: multi-document summarization, knowledge-rich approaches, current topics) Helena Ahonen-Myka Spring 2005. In this part…. Summarization of multiple documents MEAD Knowledge-rich approaches STREAK
E N D
Processing of large document collections Part 7 (Text summarization: multi-document summarization, knowledge-rich approaches, current topics) Helena Ahonen-Myka Spring 2005
In this part… • Summarization of multiple documents • MEAD • Knowledge-rich approaches • STREAK • Current topics in text summarization
Summarization of multiple documents • Radev, et al (2004): Centroid-based summarization of multiple documents • idea: summarizing news events • news stories come from several sources (e.g. news agencies) • all news stories talking about the same event (e.g. accident, earthquake,…) are clustered • stories in one cluster repeat (partially) the same content • stories have a chronological order (time stamp) • one summary for each cluster is created • a reader does not have to read the same content several times
Centroid-based clustering • each document is a tf * idf weighted vector • documents are clustered • cluster centroid = first document • a new document D is compared to each centroid C • if sim(C, D) > threshold, D is included in C, and C is updated • if D is not included in any cluster, it becomes the centroid of a new cluster
MEAD extraction algorithm • sentences are ranked according to a set of features • input: • a cluster of documents, segmented into n sentences • compression rate r • output: • a sequence of n x r sentences from the original documents • presented in the same order as in the input documents
Features • three features: • centroid value Ci for sentence Si is the sum of the centroid values of all words in the sentence • the centroid vector of the cluster represents importance of words for all the documents in the cluster
Features • positional value: • Cmax = score of the highest-ranking sentence in the document according to the centroid value • the ith sentence in a document gets a value
Features • first sentence overlap Fi: • the inner product of the current sentence Si and the first sentence of the document • combined score of sentence Si: linear combination of three features • score(Si) = wcCi + wpPi + wfFi
Cross-sentence dependencies • scores of sentences can be further refined after considering possible cross-sentence dependencies, for instance • repeated content in sentences • redundant content can be removed • chronological ordering • earlier or later sentences can be preferred • source preferences • e.g. Helsingin sanomat is trusted more than Iltalehti…
Repeated content • John Doe was found guilty of the murder. • The court found John Doe guilty of the murder of Jane Doe last August and sentenced him to life. (2. presents additional content -> 1. redundant) • Eighteen decapitated bodies have been found in a mass grave in northern Algeria, press reports said Thursday. • Algerian newspapers have reported on Thursday that 18 decapitulated bodies have been found by the authorities. (equivalent content)
Reranking based on repeated content • redundancy penalty Rij for each sentence i which overlaps with sentences j that have higher score value • redundancy penalty for a sentence: max (Rij) • new_score(si) = wcCi + wpPi + wfFi – wRRi • all sentences are reranked by new_score and a new extract in created • iteration until reranking does not result in a different extract
Knowledge-rich approaches • structured information can be used as the starting point for summarization • structured information: e.g. data and knowledge bases • may have been produced by processing input text (information extraction) • summarizer does not have to address the linguistic complexities and variability of the input, but also the structure of the input text is not available
Knowledge-rich approaches • there is a need for measures of salience and relevance that are dependent on the knowledge source • addressing cohesion, and fluency becomes the entire responsibility of the generator
STREAK • McKeown, Robin, Kukich (1995): Generating concise natural language summaries • goal: folding information from multiple facts into a single sentence using concise linguistic constructions
STREAK • produces summaries of basketball games • first creates a draft of essential facts • then uses revision rules constrained by the draft wording to add in additional facts as the text allows • revision rules have been extracted by studying human-written game summaries
STREAK • input: • a set of box scores for a basketball game • historical information (from a database) • task: • summarize the highlights of the game, underscoring their significance in the light of previous games • output: • a short summary: a few sentences
STREAK • the box score input is represented as a conceptual network that expresses relations between what were the columns and rows of the table • essential facts: the game result, its location, date and at least one final game statistic (the most remarkable statistic of a winning team player)
STREAK • essential facts can be obtained directly from the box-score • in addition, other potential facts • other notable game statistics of individual players - from box-score • game result streaks (Utah recorded its fourth straight win) - historical • extremum performances such as maximums or minimums - historical
STREAK • essential facts are always included • potential facts are included if there is space • decision on the potential facts to be included could be based on the possibility to combine the facts to the essential information in cohesive and stylistically successful ways
STREAK • given facts: • Karl Malone scored 39 points. • Karl Malone’s 39 point performance is equal to his season high • a single sentence is produced: • Karl Malone tied his season high with 39 points
Current topics in text summarization • multi-document summarization • non-extrative summarization (abstracts) • spoken language (incl. dialogue) summarization • multilingual summarization • integrating of question answering and text summarization • web-based & multimedia summarization • evaluation of summarization systems • Document Understanding Conferences (DUC)
Mapping to the information retrieval process information need documents query document representations matching result query reformulation